RECIPROCITY-BASED EMOTIONS: AN EXPERIMENTAL STUDY OF …

RECIPROCITY-BASED EMOTIONS:AN EXPERIMENTAL STUDY OF THE EFFECTS OF EXPERIENCING

POSITIVE AND NEGATIVE RECIPROCITY*

ABSTRACT

This paper presents an experimental analysis of people’s behavior in situations involving bothpositive and negative reciprocity. The experiment implements sequences of two types of extensiveform games called Punishment games and Trust games. The contemporaneous use of these twotypes of games allows us to define an ideal framework for understanding the basic elements of reci-procal behavior. Results show that the level of trust and punishment are consistent with the viewthat emotions are involved.

JEL Classification: D 63, C 78, C 91.Key words: Reciprocity, trust, intentions, emotions, experiments.

I - INTRODUCTION

This paper studies human behavior in games with underlying reciprocity respon-ses. Fehr and Gächter (2000) have defined reciprocity as follows: «Reciprocitymeans that in response to friendly actions, people are frequently much nicer andmuch more cooperative than predicted by the self-interest model; conversely, inresponse to hostile actions they are frequently much more nasty and even brutal».This definition implies two implicit concepts: positive reciprocity and negativereciprocity. People respond to friendly or hostile actions disregarding material incen-tives.

Experimental results from ultimatum bargaining games, trust games, gift exchan-ge games, public goods experiments, and social dilemma games show evidence forreciprocal behaviors (negative and positive reciprocity). In experimental literaturewe can find several explanations for the evidence of reciprocal behavior: i) inequalityaversion (Bolton - Ockenfels, 2000; Fehr - Schmidt, 1999); ii) person-based respon-se: people respond to the type of person they face (Levine, 1998); iii) intention-basedresponse: a desire to reward good intentions or punish bad intentions (Brandts - Solà,

* I would like to thank James C. Cox, Luis González Morales, Astrid Hopfensitz and RupertSausgruber for their suggestions and comments. I am particularly grateful to Samuel Bowles for hisvaluable comments and suggestions. This work was supported by Action Concertée Incitative,Systèmes Complexes en Sciences Humaines et Sociales, from CNRS.

2001; Falk - Fehr - Fischbacher, 2000); and boundedly rational behavior (Roth -Erev, 1995; Gale - Binmore - Samuelson, 1995).

Reciprocity has been modeled as evolutionarily stable behavior (Güth - Yaari,1992; Bowles - Gintis, 1999). In this sense, reciprocity represents the best strategy interms of fitness, i.e., no other strategy can enter the population and perform better.Bolton (1997) shows how a 50-50 split in a bargaining problem is a rational strategyin an evolutionary framework, i.e., it is an evolutionarily stable strategy (ESS).

Young (1993 b) theoretically indicates 50-50 split as stochastically stable divisionin the bargaining problem characterized by two bargainers randomly matched fromtwo classes of agents1. This unique stable equilibrium arises as a result of theevolutionary bargaining process when mobility between the two classes is allowed2.In this case 50-50 split is a «convention», meaning a self-enforcing and expectedpattern of behavior (Lewis, 1969; Sugden, 1986).

The theory of Reciprocal Altruism (Trivers, 1971; Axelrod - Hamilton, 1981;Axelrod, 1984) is consistent with the evolutionary approach. This was noticed byDawes and Thaler (1998): «An implication of reciprocal altruism is that individualswill be uncooperative in dilemma situations when there is no possibility of futurereciprocity from others, as in situations of anonymity or interacting with people on a“one shot” basis». But the experimental evidence is in favor of a more «radical»definition of reciprocity3: people engage in reciprocal behaviors even in singleinteractions (evidence of 40-60 percent of socially optimal contribution in «one shot»public good experiments; evidence of 50% trust and 75% reciprocal behaviors in thesingle play Trust Game by McCabe - Smith, 2000). Therefore, reciprocal altruismgives a limited explanation of the observed cooperation (and reciprocal behaviors).

Our hypothesis is that human minds have a propensity to establish long-termreputations as cooperators and non-defectors, and consequently people are willing toincur the risk that their anonymous counterparts are like-minded persons (see Cori-celli - McCabe - Smith, 2000). This hypothesis implies that some subjects in single-play extensive form games will opt, or respond to, moves in which intentions are tosignal a desire for positive reciprocity, and the achievement of greater individual as

1 See Young (1993 a) for a formal analysis of the stochastically stable equilibria. This analysisis based on graph theory. Stochastic stability is explained by the dynamics of the process ofequilibrium selection in a no-learning and no-reputation environment. The theory does not assumecommon knowledge over agents’ utility functions.

2 The stochastically stable division is the asymmetric Nash solution in the cases of homo-geneous agents (in terms of utility function and information) in both classes. Uniqueness isachieved if precision of demand tend to 0. The asymmetry arises for the difference in the amountof information. The class with more agents has more information. When both classes are charac-terized by heterogeneous agents, the stochastically stable division is still a unique asymmetricNash; but in this case the asymmetry arises for the difference in the distributions of types betweenthe two classes, and not for their relative dimension.

3 In analogy to the concept of strong reciprocity. See Gintis (2000) for an evolutionary model.

360 G. CORICELLI

well as social surplus than if each agent played opportunistically4. This perspectiveonly partially explains the pattern of trust and reciprocal behavior in our experiment.In addition, we assume that experienced emotions, related with our response toothers’ behavior and intentions, matter.

The specific emotions that are related with negative and positive reciprocity areanger and elation (see Geanakoplos - Pearce - Stacchetti, 1989). Anger (elation) isrelated with counterfactual thinking (see Lewis, 1973; Roese - Olson, 1995; Kahne-man - Miller, 1986) that arises when individuals compare the obtained outcome withbetter (worse) outcomes that might have been realized (see Niedenthal et Al., 1994;Zeelenberg et Al., 1999). We use the terms upward or downward counterfactual whenthe obtained outcome is compared with better outcomes or worse outcomes, respec-tively. Anger and elation result from upward and downward counterfactual, respecti-vely.

In the case of reciprocal interaction the obtained outcome depends on the decisionof two or more agents. We refer to anger (or elation) if individuals will feel justpartially responsible for their final outcome. Indeed, they will attribute the responsi-bility of their outcome to another individual or group of individuals5. The counterfac-tual reasoning allows individuals to focus on the alternative outcome that they wouldhave obtained if the other individual (or individuals) had behaved differently. Thisprocess is based on the attribution of intentionality to the other individual’s behavior(see Rabin, 1993; Brandts - Solà, 2001; Falk et Al., 1999) .

In the specific case of our experimental framework distinct emotions are linked tothe appraisals of responsibility for the determination of the final outcome, in accord-ance with the appraisal theory of emotion (see Frijda, 1986; Frijda et Al., 1989).«When people perceive some other person to be the cause of their misfortune, theyfeel angry; when people perceive impersonal circumstances beyond human control tobe the cause of their misfortune they feel sad; when they perceive themselves to bethe cause of their misfortune they fell guilty» (cf. Keltner et Al., 1993, p. 741).

We assume that the experience of anger and elation affects individual choice. Wename these emotions as «reciprocity-based» emotions. In particular, experiencedreciprocity-based anger increases the willingness to punish (even if this is costly),and decreases trust. Even though we did not use a direct measure of emotionalarousal, we strongly believe that our experimental results show the presence ofreciprocity-based emotions and their effects to the levels of trust and punishment.

In the next Section we illustrate the design and the procedure of our experiment;Section III briefly reports the experimental details; Section IV reports our results; andSection V concludes the paper with discussion.

4 Opportunism means that one is only interested in personal monetary payoff. Standard gametheory assume common knowledge of opportunism.

5 For consideration about responsibility see Connolly - Ordoñez (1997); and Zeelenberg et Al.(2000).

RECIPROCITY-BASED EMOTIONS 361

II - EXPERIMENTAL DESIGN AND HYPOTHESES

Subjects participate in two-player extensive form games. The games are of twotypes: Punishment Games (P)6 and Trust Games (T)7 (see Fig. 1). The subjects playtwo sequences of games. Each sequence is composed of three games of the same type(trust or punishment).

There are 6 subjects for each experimental session. The subjects are matchedfollowing the matching rules described in Figure 2. These two types of games (P andT) are introduced in order to study negative and positive reciprocity. In the puni-shment game the first player can move left, ending the game and splitting an amountof money, or move right and possibly obtain a greater payoff for herself and a smallerpayoff for the second player, if the second player does not punish her. Punishmentoccurs if the second player moves right, thus ending the game with zero payoff forboth players.

The sequence of punishment games is characterized by an increase in the differen-ce between the payoffs that the second player would have got if the first player hadplayed non-opportunistically (moving left)8, compared to the payoffs if the firstplayer moves right and the second does not punish her (moving left). These differen-ces are 26, 31, and 36 experimental dollars, for the first, second, and third punishmentgame.

In this sequence the cost of punishing (i.e., the difference between the secondplayer’s payoff of moving left and moving right) is the same (i.e., is equal to 12experimental dollars in each game). It is identical also the improvement in the firstplayer’s payoffs for moving right and the second player moving left (i.e., it is equalto 30 experimental dollars for each one of the three games of this sequence). Thesubgame perfect Nash equilibrium solution is always to move right for the first player

6 The punishment game is a truncation of an ultimatum game. In the two-person ultimatumgame the first mover proposes how to divide an amount of money, a pie. The responder can eitheraccept or reject the offer. If the second mover rejects the offer, both end up with nothing.Experimental results show a mean offer of 30-40% of the pie, and offers below 20% are almostalways rejected.

7 The trust game is a simplified version of an investment game (see Berg et Al., 1995). TheInvestment Game is a two-person sequential game; the first mover (player 1) must choose howmuch of $10 to send to the second mover (player 2). Both players know that the amount sent willbe tripled when it reaches player 2. Player 2 must then decide how much of the tripled money tosend back to player 1 (and how much to keep). The money sent back to player 1 does not tripleagain. Game theory predicts that rational agents playing once will act as follows: player 2 has adominant strategy to keep all of the tripled money sent by player 1; player 1 should infer thisbehavior by player 2, and therefore send nothing. In fact, this does not happen. Over 90% of theplayer 1s exhibit some degree of trust by sending some amount of money, and over a third of theplayer 2s show themselves to be trustworthy by sending back more than player 1 initially sent.

8 For a definition of opportunistic behavior, see supra note 5.

362 G. CORICELLI

and always to play left for the second player. Different behaviors indicate deviationfrom the standard game theoretical solution. If the first player moves left, ending thegame, she behaves non-opportunistically. If this is the case, the second player cancompare the obtained outcome with the worse outcomes she would have obtained ifthe first player had moved right (downward counterfactual). This comparison woulddetermine a positive emotion (elation) as a response to the first player’s behavior. Ifthe first player moves right, playing opportunistically, the second player wouldexperience worse outcomes. In this case, the result of upward counterfactual inducesanger.

FIG. 1 - a) Sequence of extensive form punishment games: P1, P2, and P3, respectively. b) Sequence of extensive form trust games: T1, T2, and T3, respectively.

Note: In each game player 1 moves first by playing «left» or «right». If player 1 moves «left» thegame is over. If player 1 moves «right» then player 2 can play «left» or «right», ending the game.Wherever the game ends, player 1 gets the payoff at the top, and player 2 gets the payoff at thebottom of the box.


a)

b)

38

38

1

Cooperative

Not-punish

Opportunistic

Punish

2

1

2

1

2

1

Not-trust

Reciprocate

Trust

Defect

2

1

2

1

2

68

12 0

0

43

43

73

120

0

48

48

72

12 0

0

9

9

36

47 0

81

15

15

42

550

89

18

18

45

60 0

94

The second sequence of games is composed of three trust games. In the trust gamethe first player can move left, ending the game with a small payoff for both players,or she can move right, giving the second player the opportunity to choose betweenreciprocating and defecting. If the second player chooses left, she and the first playerget more than the first possible outcome; whereas the game ends with her maximumpayoff and with a zero payoff for the first player if she chooses right. In this game ifthe first player moves right, she trusts the second player to reciprocate (i.e., not tomove right). The subgame perfect Nash equilibrium solution is always to move leftfor the first player and always to move right for the second player. In the sequence oftrust games (see Fig. 1b) the potential benefit (in terms of increasing payoffs if thesecond player reciprocates) for the first player to move right (trust) is the same for thethree games (i.e., is always equal to 27). The difference in payoffs for the secondplayer between moving right (defect) and moving left (reciprocate) is always equal to34 experimental dollars. What varies through the sequence is the difference betweenthe second player’s payoff when she moves left (reciprocate) and her payoff when thefirst player has ended the game, moving left (these differences are: 38, 40, and 42 forthe first, second and third trust game, respectively). In other words, what varies is thebenefit that the second player gets from the first player’s trusting behavior. In thisway first and second players’ patterns of behavior through the three games areexplained by their motivation in terms of reciprocity.

In the sequence of trust games the first player can experience anger from thesecond player’s defecting behavior. If the second player goes right, the first playergets a payoff of zero experimental dollars. In this case, and also in the case in whichshe cannot trust the second player (she moves left), she might compare her payoffwith a better non-obtained payoff (upward counterfactual). When the first playermoves right and second player moves left, the first player would compare her payoffwith worse payoffs (downward counterfactual). This would determine elation andappreciation of the reciprocal behavior of the second player.

The last part of the analysis focuses on the relationship between the two sequen-ces. We assume that subjects are carrying over emotional experience from onesequence to another, and different emotional experiences determine different beha-viors. For this reason we introduce two conditions. The subjects in condition 1(punishment-trust) play first the three punishment games and then the three trustgames; whereas in condition 2 (trust-punishment) subjects play first the three trustgames and then the three punishment games. The matching protocol for condition 2is described in Figure 2b. This protocol is identical to the first one with the order ofthe two sequences inverted. Figures 2a and 2b show that the position (first or secondmover) of the players in the sequences is the same, and it is alternating, i.e., thesubjects that were players one in the first sequence are players two in the secondsequence, and the opposite for the others. We formulate two behavioral hypothesesrelated to the difference between the two conditions.

364 G. CORICELLI

HYPOTHESIS 1 The proportion of punishment behavior of the second players incondition 2 is higher than in condition 1.

HYPOTHESIS 2 The proportion of trust behavior of the first players in condition 1 islower than in condition 2.

These two hypotheses are based on considerations about subjects’ emotionalexperience throughout the experiment. Having experienced opportunistic behaviorfrom the others increases negative reciprocity, and decreases trust in positive recipro-city.

III - EXPERIMENTAL DETAILS

Subjects do not know the matching protocol implemented. They know that theyare always playing with a different opponent from game to game, but they do notknow with whom they are playing (see the Instructions in the Appendix). Anonymityis maintained; subjects are paid separately and with a double blind procedure. Theexperimenter only knows the ID number of the subject for data analysis purposes, butdoes not know the corresponding name. This procedure is made clear through theinstruction. The games are not labeled. The subjects are informed on their positionbefore playing a game (e.g. first player or second player). The subjects see all thepayoff information, and players 2 see the move of players 1 before making a decision.

The subjects of our experiment were 78 undergraduate students from the Univer-sity of Arizona. 48 subjects participated in condition 1, and 30 subjects participatedin condition 2. They received 5 US dollars show up fees and two of the payoffsobtained during the experiment. The two payoffs, one from each sequence, wererandomly chosen at the end of the experiment. This procedure has been introduced inorder to avoid endowment effects. The payoffs were converted at a rate of 10Experimental Dollars to 1 US dollar. The experiment was computerized.

IV - RESULTS

4.1. Means choice proportion: Punishment games

The result of the experiment shows that 68 percent of the first players in thepunishment games had opportunistic behavior, and 25 percent of second playersresponded by punishing them. The mean proportion of opportunistic behavior of thefirst players in the punishment games does not change significantly along the sequen-ce (see Figs. 3a and 3b). This is true in both conditions (McNemar test, P1 vs. P2,p-value=0.23; P1 vs. P3, p-value=0.79; and P2 vs. P3, p-value=0.51). As shown inFigures 3a and 3b, the proportion of punishment inflicted by the second playersincreases along the sequence. In particular, there is an increase from the first to thesecond punishment game, and then there is a drop of punishment in the last game, thisproportion being anyway higher than the first game.


FIG. 2 - a) Matching Protocol. Condition 1. b) Matching Protocol. Condition 2.

4.2. Means choice proportion: Trust games

In the trust games 50 percent of the first players trusted the second players, and 45percent of them reciprocated. Figures 3c and 3d show mean choice proportions offirst and second players for the sequence of trust games in condition 1 and condition2, respectively. The level of trust significantly decreases from the first game of thesequence in both conditions (McNemar test, T1 vs. T2, p-value=0.02; T1 vs. T3,p-value=0.03), but does not vary from the second to the third game of the sequence(McNemar test, p-value=1).

Figures 3c and 3d show an increase in the proportion of second players’ reciprocalbehavior along the sequence of trust games.

366 G. CORICELLI

b)

SEQUENCE 1 SEQUENCE 2

T1 T2 T3 P1 P2 P3 Type of

Game

Position

Player 1

Player 2

Player 1

Player 2

Player 1

Player 2

ID#1

ID#2

ID#1

ID#4

ID#1

ID#6

ID#2

ID#1

ID#4

ID#1 ID#6

ID#1

ID#3

ID#4

ID#3

ID#6

ID#3

ID#2

ID#5

ID#6

ID#5

ID#2

ID#5

ID#4

ID#4

ID#3

ID#6

ID#3 ID#2

ID#3

ID#6

ID#5

ID#2

ID#5 ID#4

ID#5

SEQUENCE 1 SEQUENCE 2

P1 P2 P3 T1 T2 T3 Type of

Game

Position

Player 1

Player 2

Player 1

Player 2

Player 1

Player 2

ID#1

ID#2 ID#1

ID#4

ID#1

ID#6

ID#2

ID#1

ID#4

ID#1

ID#6

ID#1

ID#3

ID#4 ID#3

ID#6

ID#3

ID#2

ID#5

ID#6 ID#5

ID#2

ID#5

ID#4

ID#4

ID#3

ID#6

ID#3

ID#2

ID#3

ID#6

ID#5

ID#2

ID#5

ID#4

ID#5

a)

FIG. 3 - Proportion of behavior in condition 1 and condition 2 for each sequence of threepunishment and trust games: a) Opportunistic and punishment behavior in condition 1, and

b) in condition 2; c) Trust and reciprocity in condition 1, and d) in condition 2.

4.3. Interaction between first and second players’ behavior

Table I presents a contingency table with the first players’ behavior (rows) in thepunishment game conditioned to the second players’ earlier moves (columns). Thechi-square test shows that we can reject (p-value=0.0025) the hypothesis of inde-pendence between rows and columns of this contingency table, i.e. first players’behavior depends on second players’ past (t-1) behavior9. In particular the first

9 The same test over the first players’ behavior in the punishment game conditional to secondplayers’ past (t-2) behavior is not significant (Chi-square=1.28, p-value=0.25).


c) Trust and reciprocity in condition 1. d) Trust and reciprocity in condition 2.

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

T1 T2 T3

pro

port

ion

t rust

reciprocity

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

T1 T2 T3

pro

port

ion

t rust

reciprocity

a) Opportunistic and punishment b) Opportunistic and punishment behavior in condition 1. behavior in condition 2.

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

P1 P2 P3

pro

po

rtio

n

opport

punish

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

P1 P2 P3

pro

port

ion

opport

punish

players are responding to the second players’ punishment behavior. Experiencingpunishment significantly decreases opportunistic behavior.

TABLE I

Play

ers

1

Players 2

Not-punish (t-1) Punish (t-1)

Not-opportunistic (t) 3 6

Opportunistic (t) 38 6

Chi-square=9.16p-value=0.0025

Contingency table. The entries are numbers of subjects. Punishment games. First players’ behavior(rows) conditional to past (t-1) second players’ behavior (columns).

Note: Not-opportunistic=player 1 moves left at time t, opportunistic=player 1 moves right at t;not-punish=player 2 moved left at t-1, punish=player 2 moved right at t-1.

Table II shows the result of the second players’ behavior (rows) contingent to firstplayers’ behavior at time t-1 (columns). The second players responded to firstplayers’ opportunistic behavior (chi-square=8.71, and p-value=0.003). In particularthere is a significant increase in punishment as a response to past opportunisticbehavior. It is interesting to notice that second movers never punished if they had notexperienced opportunistic behavior at time t-1.

TABLE II

Play

er2

Players 1

Not-opportunistic (t-1) Opportunistic (t-1)

Not-punish (t) 18 20

Punish (t) 0 15


Contingency table. The entries are numbers of subjects. Punishment games. Second players’behavior (rows) conditional to past (t-1) first players’ behavior (columns).

Note: Not-punish=first player moves left at time t, punish=first player moves right at t; not-opportun-istic=first player moved left at time t-1, opportunistic=first player moved right at t-1.

368 G. CORICELLI

Table III shows the behavior of first players in the trust games. The contingencytable represents first players’ trusting behavior (rows) contingent to second players’reciprocal behavior at time t-1 (columns). We reject the hypothesis of independencebetween rows and columns (p-value=0.013), i.e. trust depends on earlier secondmovers’ reciprocal behavior10.

TABLE III

Play

ers

1

Players 2

Reciprocate (t-1) Defect (t-1)

Not-trust (t) 2 11

Trust (t) 17 10


Contingency table. The entries are numbers of subjects. Trust games. First players’ behavior (rows) conditional to past (t-1) second players’ behavior (columns).

Note: Not-trust=first player moves left at time t, trust=first player moves right at time t; recipro-cate=second player moved left at time t-1, defect=second player moved right at time t-1.

Table IV illustrates second players’ behavior in the trust game. The chi-squaretest cannot reject (p-value = 0.185) the hypothesis of independence between the rows(second players’ reciprocal behavior at time t) and the columns (first players’ trustingbehavior at time t-1). This indicates that second players in the trust games were nomore likely to reciprocate if they have been trusted in earlier period.

TABLE IV

Play

ers

2

Players 1

Not-trust (t-1) Trust (t-1)

Reciprocate (t) 4 13

Defect (t) 7 6


Contingency table. The entries are numbers of subjects. Trust games. Second players’ behavior(rows) conditional to past (t-1) first players’ behavior (columns).

Note: Reciprocate=second player moves left at time t, defect=second player moves left at time t;not-trusted=first player moved left at time t-1, trusted=first player moved right at time t-1.

10 The same test over the first players’ behavior in the trust game conditional to second players’past (t-2) behavior is not significant (Chi-square=1.42, p-value=0.23).


4.4. Condition 1 vs. Condition 2. Support for hypotheses 1 and 2: Sequence matters

Figure 4 plots subjects’ behavior for the two conditions. Wilcoxon rank-sum testfor differences of means of subjects’ behavior in condition 1 and condition 2 showsno significant difference between the mean proportions of first players opportunisticbehavior in the punishment games of condition 1 and condition 2 (p-value=0.79).Punishment in condition 2 is higher than in condition 1 (p-value=0.09), this result isin support of hypothesis 1. Trust is significantly higher in condition 2 than incondition 1 (p-value=0.049), in support of hypothesis 2. Finally, there is no signifi-cant difference between reciprocity in condition 2 and condition 1 (p-value=0.218).These results show that having experienced a sequence of punishment games assecond player or a sequence of trust games as first player affect behavior. In particu-lar, it increases subject’s sensitivity to other subjects’ behavior. They tend to attributeto the others more responsibility and intentionality for their actions.

FIG. 4 - Proportions of behavior in condition 1 where subjects play the sequence of threepunishment games first and then the sequence of three trust games; and condition 2, where the

order of the sequences is reversed.

370 G. CORICELLI

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

Opportunistic Punishment Trust Reciprocity

pro

po

rtio

n

condtion 1

condition 2

4.5. Test for alternative explanations: Efficiency and inequality aversion

Our design allows us to distinguish consistent patterns of behavior in terms ofefficiency (desire to maximize the total payoffs)11 and inequality aversion12. Asshown from the individual data (see Appendix A2, and A3), there is no significantevidence of consistent patterns of behavior motivated by efficiency and inequalityaversion. Only five subjects followed the «efficiency pattern» in both sequences (ID#39, 41, and 44, in condition 1; ID# 3, and 13, in condition 2), and only one subjectbehaved consistently with a pattern of inequality aversion (ID# 32, in condition 1).We report in Appendix A4 the relative frequencies of patterns of behavior in thesequence of punishment and trust games, respectively. The patterns of behaviormotivated by efficiency or inequality aversion are never relatively more importantthan other behavior patterns13.

V - DISCUSSION

In our experiment subjects played two sequences of two-person extensive formgames with anonymous and different counterparts. The payoffs structure of thegames determined underlying reciprocity responses. The anonymity and the chan-ging of counterparts avoided reputation effects. This procedure should have inducedthe subjects to play without changing their behavior throughout the games, as if theywere playing a one-shot game each time. This was not the case. Results showsubjects’ sensitivity to the change (introduced by the experimental design) of thelevels of the determinants of the «reciprocal interaction». In the sequence of puni-shment games we increased through the three games the level of the potential «pain»inflicted by the first movers to the second movers. In the sequence of trust games weincreased through the three games the second movers’ benefits determined by thefirst movers’ trusting behavior. Results show subjects’ sensitivity to these changes.The second players in the punishment games punished more through the sequence;and the second players in the trust games defected less through the sequence. Thiscorroborates previous findings by Cooper and Kraker Stockman (2002) that: subjectslearn to punish over time. Subjects do not show fixed preferences through thesequences of our experiment. Indeed, subjects’ preferences depend on other players’behavior, and on the interpretation of this behavior in terms of their intentions (inanalogy to the findings of Brandts - Solà, 2001; and Falk et Al., 1999).

11 See Charness - Rabin (2002).12 We have the same pattern for the two theories on inequality aversion: Bolton - Ockenfels

(2000), and Fehr - Schmidt (1999).13 With the exception of the case of second players in the punishment games in condition 1, in

which the relative frequency of efficiency based behavior is 0.68.


The second part of our analysis is dedicated to the study of different behaviorsbetween the two conditions of our experimental design. The only difference betweencondition 1 and condition 2 is the order of the sequences played, i.e., in condition 1the subjects first played the three punishment games and then the three trust games,whereas, in condition 2 the subjects first played the three trust games and then thethree punishment games. The order and the payoffs structure of the games inside eachsequence is the same in both conditions. Results show that the subjects (secondplayers) in condition 2 punished more than the subjects in condition 1, and that thesubjects in condition 2 trusted more than the subjects in condition 1. We concludethat different experiences determine different behavior. The second players in condi-tion 1 experienced a sequence of increased pain (inflicted by the first players), beforethey became first players in the sequence of trust games. They trusted less than thefirst players in condition 2. The first players that started with the sequence of trustgames experienced a high level of defection before they became second players in thesequence of punishment games. They punished more than the second players in thepunishment games in condition 1. The negative emotion caused by having experien-ced defection (from others) is carried over from one sequence to another. Subjectscompare their obtained outcomes with the outcome they would have obtained if theircounterpart behaved non-opportunistically. This comparison, based on upward coun-terfactual, determines «negative» emotional response that «influences on subsequentjudgments, making the role of other people in causing negative events salient andperhaps events in general» (cf. Keltner et Al., 1993, p. 751). We named this responseas reciprocity-based anger. Experiencing anger determines punishment and non-tru-sting behavior.

Negative reciprocity is triggered by emotional response to others’ behavior. Inthis sense emotion contributes to the enforcement of social norms (see Gintis, 2000;Bowles - Gintis, 2001; Fehr - Gächter, 2002; Carpenter et Al., 2002)14. Our resultsshow that subjects are willing to enforce the social norm of cooperation even if thisprocess is based on costly punishment15.

We conclude that different behavior is determined by different emotional expe-rience. Indeed, experienced emotions matter in reciprocal interactions. As stated byBowles and Gintis (2001): «Adherence to social norms is underwritten by emotions,not only by the expectation of future reciprocation».

14 We adopt the definition of social norm given by Fehr - Gächter (2000 b): «It is: 1) abehavioral regularity; that is 2) based on a socially shared belief of how one ought to behave; whichtriggers 3) the enforcement of the prescribed behavior by informal social sanctions» (cf. p. 9).

15 It is noteworthy to say that we found less responsiveness to the experience of being trusted,i.e. having being trusted in the past does not influence current reciprocal (trustworthy) behavior.

GIORGIO CORICELLIInstitute des Sciences Cognitives

CNRS, Lyon

372 G. CORICELLI

A p p e n d i x A 1

Instructions

In this experiment you will participate in a two-person decision task. This task will be repeated a number of times, called decision periods.Each decision period you will be paired with another individual: your counterpart. You

will always play with a different counterpart from game to game, but you will not know withwhom you are playing. The joint decisions made by you and your counterpart will determinehow much money you might earn for that decision period. At the end of the experiment twoof the decision period will be randomly chosen and you will be paid according with thecorresponding outcomes. Your earnings from the chosen decision periods will be paid to youin cash at the end of the experiment. We will not tell anyone else your earnings and we askthat you do not discuss your earnings.

You and your counterpart will participate in the decision task shown below. In the courseof the task you will be player 1 or 2. This information will be shown to you in an informationscreen before the decision task.

Notice the open boxes with numbers in them. They show the different earnings that youand your counterpart can make. The amount on top is your earning if this box is reached. Theamount on the bottom is the other player’s earning.

You and your counterpart will jointly determine a path through the diagram to anearnings box. We will highlight the box when you reach it.

A path is defined to be a sequence of moves through the diagram. A move is defined as achoice of direction in the diagram. The highlighted lines show the possible directions ofmoves that can be made. Moves are always left or right.

The picture indicates which player gets to move at that point in the diagram. Let’s try asimple example to see how the task works.

If you have questions, please ask them now. Otherwise get ready to play.


A p p e n d i x A 2

Individual data. Condition 1

Note: L=left, R=right, player 1. l=left, r=right, player 2. X=no observation.

374 G. CORICELLI

Session 1 Session 2

ID Position P1 P2 P3 Position T1 T2 T3

1 1 R R R 2 r X r

2 1 L L L 2 X l l

3 1 R R R 2 l X X

4 1 L R R 2 X X r

5 1 R R R 2 r X X

6 1 L R R 2 X X X

7 1 L R R 2 r l X

8 1 R R L 2 X X X

9 1 L L L 2 r r l

10 1 L R R 2 l l X

11 1 L R L 2 X X X

12 1 R R R 2 X r X

13 1 L L L 2 X X l

14 1 R R R 2 X X X

15 1 R R R 2 r X X

16 1 R R L 2 r X X

17 1 R R L 2 l X r

18 1 L L R 2 r r r

19 1 R R R 2 r X X

20 1 R R R 2 X r X

21 1 R R R 2 X X X

22 1 R R R 2 X X l

23 1 L L L 2 r X X

24 1 R R R 2 X X X

25 2 l r X 1 R L R

26 2 X X r 1 L L L

27 2 l r X 1 R R L

28 2 X l l 1 L L L

29 2 l l X 1 R R R

30 2 X l l 1 L L L

31 2 X X l 1 R L L

32 2 r r X 1 L L L

33 2 X l X 1 R L L

34 2 X X l 1 R R R

35 2 X l l 1 L L R

36 2 l l l 1 L L L

37 2 X l r 1 L L L

38 2 l l l 1 L L L

39 2 l l X 1 R R R

40 2 l l l 1 L L L

41 2 l X l 1 R R R

42 2 X l X 1 R R R

43 2 l l r 1 R L L

44 2 l l l 1 R R R

45 2 l X l 1 L L L

46 2 l l l 1 L L L

47 2 X l l 1 L L L

48 2 l r X 1 R L L

A p p e n d i x A 3

Individual data. Condition 2

Note: L=left, R=right, player 1. l=left, r=right, player 2. X=no observation.


Sequence 1 Sequence 2

ID Position T1 T2 T3 Position P1 P2 P3

1 1 R R R 2 X X l

2 1 R L L 2 r r l

3 1 R R R 2 l X l

4 1 L L R 2 l X l

5 1 R R R 2 l r X

6 1 R L L 2 X l r

7 1 R R L 2 l r X

8 1 R R R 2 X l r

9 1 L L L 2 r l r

10 1 R R L 2 l r r

11 1 R L L 2 l r X

12 1 R L L 2 l X l

13 1 R R R 2 X l l

14 1 R R L 2 l l X

15 1 L R R 2 X X X

16 2 l X X 1 L R R

17 2 r X X 1 R L L

18 2 l l l 1 R R R

19 2 X r X 1 R R R

20 2 r l X 1 R L R

21 2 r l X 1 L L L

22 2 l X X 1 R R R

23 2 l l l 1 L L R

24 2 X X X 1 R L L

25 2 r l r 1 R R L

26 2 r X l 1 R R R

27 2 r r X 1 R R R

28 2 l l r 1 L R R

29 2 r X r 1 R R R

30 2 X l l 1 L R L

Appendix A 4

Relative frequencies of different patterns of behaviour

Note: E.= Efficiency, I.A.= Inequality Aversion. L=left, R=right, player 1. l=left, r=right, player 2.X=no observation.

376 G. CORICELLI

Sequence Position Patterns of behavior Condition 1

(N=48)

Condition 2

(N=30) Total

player 1 E. R L L 0/24=0 2/15=0.13 2/24=0.08

Punishment

games I.A. L R R 2/24=0.08 1/15=0.07 3/39=0.08

player 2 E. l l l, l l X, l X l, X l l 13/19=0.68 5/13=0.38 18/32=0.56

I.A. r r r, r r X, r X r, X r r 1/19=0.05 0/13=0 1/32=0.03

other

r l l, l r l, l l r, r l X, r X l,

X r l, l r X, l X r, X l r 5/19=0.26 8/13=0.62 13/32=0.41

player 1 E. R R R 6/24=0.25 5/15=0.33 11/39=0.28

Trust games I.A. L L L 11/24=0.46 0/15=0 11/39=0.28

player 2

E. or

I.A. l l l, l l X, l X l, X l l 2/7=0.29 3/10=30 5/17=0.29

other

r l l, l r l, l l r, r l X, r X l, X r l, l r X, l X r, X l r, r r r,

r r X, r X r, X r r 5/7=0.71 7/10= 0.7 12/17=0.71

R E F E R E N C E S

K. ABBINK - G.E. BOLTON - A. SADRIEH - F.-F. TANG, Adaptive Learning versus Punishment inUltimatum Bargaining, in «Games and Economic Behavior», 37, 2001, pp. 1-25.

J. ANDREONI, Why Free Ride? Strategies and Learning in Public Goods Experiments, Universityof Wisconsin, Department of Economics, «Discussion Paper», 1987.

R. AXELROD, The Evolution of Cooperation, Basic Books, New York 1984.

R. AXELROD - W.D. HAMILTON, The Evolution of Cooperation in Biological Systems, in «Scien-ce», 211, 1981, pp. 1390-1396.

J. BERG - J. DICKHAUT - K.A. MCCABE, Trust, Reciprocity, and Social History, in «Games andEconomic Behavior», 10, 1995, pp. 122-142.

G. BOLTON, The Rationality of Splitting Equally, in «Journal of Economic Behavior andOrganization», XXXII, 1997, pp. 365-381.

G. BOLTON - J. BRANDT - A. OCKENFELS, Measuring Motivation for the Reciprocal ResponsesObserved in a Simple Dilemma Game, in «Experimental Economics», 1, 1998, pp. 207-219.

G.E. BOLTON - A. OCKENFELS, ERC-A Theory of Equity, Reciprocity, and Competition, in«American Economic Review», XC, 2000, pp. 166-193.

S. BOWLES - H. GINTIS, The Evolution of Strong Reciprocity, University of Michigan Press, AnnArbor 1999 (mimeo).

S. BOWLES - H. GINTIS, The Economics of Shame and Punishment, «Working Papers», 2001.

J. BRANDTS - C. SOLÀ, Reference Points and Negative Reciprocity in Simple Sequential Games, in«Games and Economic Behavior», 36, 2001, pp. 138-157.

G. CHARNESS - M. RABIN, Understanding Social Preferences with Simple Tests, in «QuarterlyJournal of Economics», CXVII, 2002, pp. 817-869.

J.P. CARPENTER - P.H. MATTHEWS - O. ONG’ONG’A, Why Punish? Social Reciprocity and theEnforcement of Prosocial Norms, 2002 (mimeo).

T. CONNOLLY - L.D. ORDOÑEZ - R. COUGHLAN, Regret and Responsibility in the Evaluation ofDecision Outcomes, in «Organizational Behavior and Human Decision Processes», LXX,1997, pp. 73-85.

D.J. COOPER - C. KRAKER STOCKMAN, Learning to Punish: Experimental Evidence from aSequential Step-Level Public Good Game, in «Experimental Economics», 5, 2002, pp. 39-51.

G. CORICELLI - K.A. MCCABE - V.L. SMITH, Theory-of-mind Mechanism in Personal Exchange,in G. HATANO et Al. (ed.), Affective Mind, Elsevier Science Publisher, Amsterdam 2000, pp.250-259.

R.M. DAWES - R. THALER, Anomalies Cooperation, in «Journal of Economic Perspectives», 2,1988, pp. 187-197.


A. FALK - E. FEHR - U. FISCHBACHER, On the Nature of Fair Behavior, Institute for EmpiricalResearch in Economics, University of Zurich, «Working Paper», Zurich 1999.

A. FALK - E. FEHR - U. FISCHBACHER, Testing Theories of Fairness Intentions Matter, Universityof Zurich, «Working Paper», Zurich 2000.

E. FEHR - S. GÄCHTER, Fairness and Retaliation: The Economics of Reciprocity, in «Journal ofEconomic Perspectives», XIV, 2000a, pp. 159-181.

E. FEHR - S. GÄCHTER, Cooperation and Punishment in Public Goods Experiments, in «AmericanEconomic Review», XC, 2000b, pp. 980-994.

E. FEHR - S. GÄCHTER, Altruistic Punishment in Humans, in «Nature», 415, 2002, pp. 137-140.

E. FEHR - K. SCHMIDT, A Theory of Fairness, Competition, and Cooperation, in «QuarterlyJournal of Economics», CXIV, 1999, pp. 817-868.

N.H. FRIJDA, The Emotions, Cambridge University Press, Cambridge 1986.

N.H. FRIJDA - P. KUIPERS - E. SCHURE, Relations among Emotion, Appraisal, and EmotionalAction Readiness, in «Journal of Personality and Social Psychology», LVII, 1989, pp. 212-228.

J. GALE - K.G. BINMORE - L. SAMUELSON, Learning to be Imperfect: The Ultimatum Game, in«Games and Economic Behavior», 8, 1995, pp. 56-90.

J. GEANAKOPLOS - D. PEARCE - E. STACCHETTI, Psychological Games and Sequential Rationality,in «Games and Economic Behavior», 1, 1989, pp. 60-79.

H. GINTIS, Strong Reciprocity and Human Sociality, in «Journal of Theoretical Biology», 206,2000, pp. 169-179.

W. GÜTH - M. YAARI, An Evolutionary Approach to Explain Reciprocal Behavior in a SimpleStrategic Game, In U. WITT (ed.), Explaining Process and Change-Approaches to EvolutionaryEconomics, University of Michigan Press, Ann Arbor 1992.

D. KAHNEMAN - D. MILLER, Norm Theory: Comparing Reality to its Alternatives, in «Psycholo-gical Review», XCIII, 1986, pp. 136-153.

D. KELTNER - P.C. ELLSWORTH - K. EDWARDS, Beyond Simple Pessimism: Effect of Sadnessand Anger on Social Perception, in «Journal of Personality and Social Perception, in «Journalof Personality and Social Psychology», LXIV, 1933, pp. 740-752.

D. LEVINE, Modeling Altruism and Spitefulness in Experiments, in «Review of EconomicDynamics», 1, 1998, pp. 593-622.

D. LEWIS, Convention: A Philosophical Study, Rutgers University Press, Cambridge (Mass.)1969.

D. LEWIS, Counterfactuals Harvard, University Press, Cambridge (Mass.) 1973.

G. LOOMES - R. SUGDEN, Regret Theory: An Alternative Theory of Rational Choice underUncertainty, in «The Economic Journal», XCII, 1982, pp. 805-824.

G. LOOMES - R. SUGDEN, Disappointment and Dynamic Consistency in Choice under Uncertainty,in «Review of Economic Studies», LIII, 1986, pp. 271-282.

378 G. CORICELLI

K.A. MCCABE - V.L. SMITH, A Comparison of Naïve and Sophisticated Subject Behavior withGame Theoretic Predictions, in «Proceedings National Academy of Sciences», XCVII, 2000,pp. 3777-3781.

J. METCALFE - W. MISCHEL, A Hot/cool-system Analysis of Delay of Gratification: Dynamics ofWillpower, in «Psychological Review», CVI, 1999, pp. 3-19.

P.M. NIEDENTHAL - J.P. TANGNEZ - I. GAVANSKI, «If only I weren’t» versus «if only I hadn’t»:Distinguishing Shame and Guilt in Counterfactual Thinking, in «Journal of Personality andSocial Psychology», LXVII, 1994, pp. 585-595.

M. RABIN, Incorporating Fairness into Game Theory and Economics, in «American EconomicReview», LXXXIII, 1993, pp. 1281-1302.

N.J. ROESE - J.M. OLSON, What Might Have Been: The Social Psychology of CounterfactualThinking, Erlbaum, Mahwah (NJ) 1995.

A.E. ROTH - I. EREV, Learning in Extensive-form Games: Experimental Data and Simple DynamicModels in the Intermediate Term, in «Games and Economic Behavior», 8, 1995, pp. 164-212.

T. SCHELLING, Self-Command in Practice, in Policy, and in a Theory of Rational Choice, in«American Economic Review», LXXIV, 1984, pp. 1-11.

R. SUGDEN, The Evolution of Rights, Cooperation and Welfare, Basil Blackwell, New York1986.

R.L. TRIVERS, The Evolution of Reciprocal Altruism, in «Quarterly Review of Biology», XLVI,1971, pp. 35-57.

H.P. YOUNG, The evolution of conventions, in «Econometrica», LXI, 1993a, pp. 57-84.

H.P. YOUNG, An Evolutionary Model of Bargaining, in «Journal of Economic Theory», LIX,1993b, pp. 145-168.

M. ZEELENBERG, Anticipated Regret, Expected Feedback and Behavioral Decision Making, in«Journal of Behavioral Decision Making», XII, 1999, pp. 93-106.

M. ZEELENBERG - W.W. VAN DIJK - A.S.R. MANSTEAD, Regret and Responsibility Resolved?Evaluating Ordoñez and Connolly’s (2000) Conclusions, in «Organizational Behavior andHuman Decision Processes», LXXXI, 2000, pp. 143-154.


Documents

RECIPROCITY-BASED EMOTIONS: AN EXPERIMENTAL STUDY OF …